Buy a NVIDIA graphic card for a desktop CUDA
Dear all,
so far i've used CUDA with my notebook, but I'm going to buy a Desktop soon. I am a student and surely i can't buy a NVIDIA Tesla GPU, nevertheless I would buy a good NVIDIA Geforce graphic card. Have you any suggestions? Currently I've seen these three graphic cards: GTX 560i, GTX 570, GTX 580. One of these might fit? Could be there some problems using these graphic cards with linux?

Thank you in advance.

Giordano
Dear all,

so far i've used CUDA with my notebook, but I'm going to buy a Desktop soon. I am a student and surely i can't buy a NVIDIA Tesla GPU, nevertheless I would buy a good NVIDIA Geforce graphic card. Have you any suggestions? Currently I've seen these three graphic cards: GTX 560i, GTX 570, GTX 580. One of these might fit? Could be there some problems using these graphic cards with linux?



Thank you in advance.



Giordano

#1
Posted 03/11/2012 12:43 PM   
Noise maybe.
Noise maybe.

#2
Posted 03/11/2012 01:15 PM   
Hi,
I don't see any problem using those cards on Linux (even if I haven't test them all). However, I would recommend not going for an over-clocked card if the intent is to run CUDA codes. I procured a GTX580 factory-OC (with a ridiculous OC factor of 3% IIRC) with which I had many issues: the memory quickly started to get erroneous. An memcheck reported about half of it erroneous. When down-clocking the card to its recommended NVIDIA frequency, everything was fine. But unfortunately, I never found the way of down-clocking the card on Linux, only on Windows. So I had a perfectly functional video card for video rendering on both Linux and Windows (the memory errors weren't having any visible effect on the screen), but for which proper CUDA computation was only possible down-clocked and thereafter on Windows. I had do just remove the card (that now lies unused on a shelf) for installing an usable one.
Hi,

I don't see any problem using those cards on Linux (even if I haven't test them all). However, I would recommend not going for an over-clocked card if the intent is to run CUDA codes. I procured a GTX580 factory-OC (with a ridiculous OC factor of 3% IIRC) with which I had many issues: the memory quickly started to get erroneous. An memcheck reported about half of it erroneous. When down-clocking the card to its recommended NVIDIA frequency, everything was fine. But unfortunately, I never found the way of down-clocking the card on Linux, only on Windows. So I had a perfectly functional video card for video rendering on both Linux and Windows (the memory errors weren't having any visible effect on the screen), but for which proper CUDA computation was only possible down-clocked and thereafter on Windows. I had do just remove the card (that now lies unused on a shelf) for installing an usable one.

#3
Posted 03/12/2012 09:08 AM   
Scroll To Top